{
  "id": "llm_local",
  "name": "LLM (local, Qwen3-30B)",
  "category": "text",
  "description": "Prompt → text via Qwen3-30B-A3B-4bit on MLX (~17GB, MoE with 3B active params = fast on big unified memory). Batch NPC dialogue, item descriptions, lore — pipe straight into tts_local. For quick/cheap calls use the m4pro Ollama endpoint instead; this is the heavyweight. Primary-node only.",
  "accepts": [],
  "produces": ["text"],
  "resources": "gpu",
  "entry": "run.py",
  "python": "venvs/llm/bin/python",
  "params_schema": {
    "type": "object",
    "properties": {
      "prompt": {"type": "string", "default": "", "description": "The user prompt"},
      "system": {"type": "string", "default": "", "description": "Optional system prompt (character sheet, style rules)"},
      "max_tokens": {"type": "integer", "default": 1024, "minimum": 16, "maximum": 8192},
      "temperature": {"type": "number", "default": 0.7, "minimum": 0, "maximum": 2}
    }
  }
}